Setup

library(tidyverse)
library(magrittr)
library(parallel)
library(pander)
library(here)
library(scales)
library(ggpubr)
library(kableExtra)
library(edgeR)
library(DT)
library(ggrepel)
library(pheatmap)
library(ggdendro)

if (interactive()) setwd(here::here())
theme_set(theme_bw())
cores <- detectCores() - 1

load(
  here::here("1_Psen2S4Ter/R/output/1_1_DE.RData")
)

k-mer analysis

jellyfish v2.3.0 was used to count kmers of trimmed fastq files that had been filtered for rRNA sequences. This was performed for 5 values: \(k = 5, 6, 7, 8, 9, 10\). Lower values of \(k\) lose specificity in comparison to higher values, however as \(k\) increases, the exponential increase of possible kmers causes limitations due to computational processing time.

k = 5

Counts

k5files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k5", pattern = "_dumps.txt", full.names = TRUE)
k5counts <- lapply(k5files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k5dge <- k5counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k5dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k5dge$samples$group <- colnames(k5dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k5dist <- k5dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 5-mers")

k5labels <- k5dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k5heat <- k5dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "5-mer counts heatmap", labels_row = k5labels)
k5heat$tree_row$labels <- k5labels
k5den <- ggdendrogram(k5heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 5-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k5pca <- k5dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k5pca)$importance %>% pander(split.tables = Inf)

k5pcaPlot <- k5pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k5dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k5pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k5pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 5"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k5pcaRrna <- k5pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k5dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k5pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 5"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k5design <- model.matrix(~rRNA, data = k5dge$samples)
k5voom <- voom(k5dge, design = k5design)
k5fit <- lmFit(k5voom, design = k5design)
k5eBayes <- eBayes(k5fit)

k5topTable <- k5eBayes %>%
  topTable(coef = colnames(k5design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

k = 6

Counts

k6files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k6", pattern = "_dumps.txt", full.names = TRUE)
k6counts <- lapply(k6files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k6dge <- k6counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k6dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k6dge$samples$group <- colnames(k6dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k6dist <- k6dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 6-mers")

k6labels <- k6dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k6heat <- k6dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "6-mer counts heatmap", labels_row = k6labels)
k6heat$tree_row$labels <- k6labels
k6den <- ggdendrogram(k6heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 6-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k6pca <- k6dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k6pca)$importance %>% pander(split.tables = Inf)

# Plot PCA
k6pcaPlot <- k6pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k6dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k6pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k6pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 6"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k6pcaRrna <- k6pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k6dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k6pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 6"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k6design <- model.matrix(~rRNA, data = k6dge$samples)
k6voom <- voom(k6dge, design = k6design)
k6fit <- lmFit(k6voom, design = k6design)
k6eBayes <- eBayes(k6fit)

k6topTable <- k6eBayes %>%
  topTable(coef = colnames(k6design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

k = 7

Counts

k7files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k7", pattern = "_dumps.txt", full.names = TRUE)
k7counts <- lapply(k7files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k7dge <- k7counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k7dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k7dge$samples$group <- colnames(k7dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k7dist <- k7dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 7-mers")

k7labels <- k7dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k7heat <- k7dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "7-mer counts heatmap", labels_row = k7labels)
k7heat$tree_row$labels <- k7labels
k7den <- ggdendrogram(k7heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 7-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k7pca <- k7dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k7pca)$importance %>% pander(split.tables = Inf)

# Plot PCA
k7pcaPlot <- k7pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k7dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k7pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k7pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 7"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k7pcaRrna <- k7pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k7dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k7pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 7"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k7design <- model.matrix(~rRNA, data = k7dge$samples)
k7voom <- voom(k7dge, design = k7design)
k7fit <- lmFit(k7voom, design = k7design)
k7eBayes <- eBayes(k7fit)

k7topTable <- k7eBayes %>%
  topTable(coef = colnames(k7design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

k = 8

Counts

k8files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k8", pattern = "_dumps.txt", full.names = TRUE)
k8counts <- lapply(k8files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k8dge <- k8counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k8dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k8dge$samples$group <- colnames(k8dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k8dist <- k8dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 8-mers")

k8labels <- k8dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k8heat <- k8dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "8-mer counts heatmap", labels_row = k8labels)
k8heat$tree_row$labels <- k8labels
k8den <- ggdendrogram(k8heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 8-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k8pca <- k8dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k8pca)$importance %>% pander(split.tables = Inf)

# Plot PCA
k8pcaPlot <- k8pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k8dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k8pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k8pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 8"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k8pcaRrna <- k8pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k8dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k8pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 8"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k8design <- model.matrix(~rRNA, data = k8dge$samples)
k8voom <- voom(k8dge, design = k8design)
k8fit <- lmFit(k8voom, design = k8design)
k8eBayes <- eBayes(k8fit)

k8topTable <- k8eBayes %>%
  topTable(coef = colnames(k8design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

k = 9

Counts

k9files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k9", pattern = "_dumps.txt", full.names = TRUE)
k9counts <- lapply(k9files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k9dge <- k9counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k9dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k9dge$samples$group <- colnames(k9dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k9dist <- k9dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 9-mers")

k9labels <- k9dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k9heat <- k9dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "9-mer counts heatmap", labels_row = k9labels)
k9heat$tree_row$labels <- k9labels
k9den <- ggdendrogram(k9heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 9-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k9pca <- k9dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k9pca)$importance %>% pander(split.tables = Inf)

# Plot PCA
k9pcaPlot <- k9pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k9dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k9pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k9pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 9"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k9pcaRrna <- k9pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k9dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k9pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 9"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k9design <- model.matrix(~rRNA, data = k9dge$samples)
k9voom <- voom(k9dge, design = k9design)
k9fit <- lmFit(k9voom, design = k9design)
k9eBayes <- eBayes(k9fit)

k9topTable <- k9eBayes %>%
  topTable(coef = colnames(k9design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

k = 10

Counts

k10files <- list.files("/hpcfs/users/a1647910/20200310_rRNADepletion/1_Psen2S4Ter/6_jellyfish2pass/k10", pattern = "_dumps.txt", full.names = TRUE)
k10counts <- lapply(k10files, function(x){
  read_delim(x, col_names = c("mer", basename(x)), delim = " ") %>%
    set_colnames(str_remove_all(colnames(.), "_6month_F3|[0-9]*_Ps2Ex3M1_|_dumps\\.txt"))
}) %>%
  purrr::reduce(full_join) %>%
  dplyr::select(mer, contains(c("WT", "Heter")))

k10dge <- k10counts %>%
  as.data.frame() %>%
  column_to_rownames("mer") %>%
  DGEList() %>%
  calcNormFactors()
k10dge$samples %<>%
  rownames_to_column("rowname") %>%
  mutate(sample = rowname) %>%
  left_join(addInfo) %>%
  column_to_rownames("rowname")
k10dge$samples$group <- colnames(k10dge) %>%
  str_extract("(WT|Heter)") %>%
  factor(levels = c("WT", "Heter"))

Properties

k10dist <- k10dge %>%
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  pivot_longer(everything(), names_to = "sample", values_to = "count") %>%
  ggplot(aes(x=count, colour = sample)) +
  geom_density() +
  labs(x = "intensity", title = "Distribution of 10-mers")

k10labels <- k10dge$samples %>% 
  mutate(label = paste0(sample, "\n", percent(rRNA, accuracy = 0.01), " rRNA")) %>% 
  .$label
k10heat <- k10dge %>% 
  cpm(log = TRUE) %>%
  as.data.frame() %>%
  t() %>%
  pheatmap(silent = TRUE, cluster_cols = FALSE,
           show_colnames = FALSE, fontsize = 9,
           fontsize_row = 10, border_color = NA,
           main = "10-mer counts heatmap", labels_row = k10labels)
k10heat$tree_row$labels <- k10labels
k10den <- ggdendrogram(k10heat$tree_row, rotate = TRUE) +
  labs(title = "Hierarchical clustering of 10-mer counts") +
  theme(plot.title = element_text(size = 12))

# Assess cpm values to make sure PCA results are not heavily skewed by highly expressed genes
k10pca <- k10dge %>%
  cpm(log = TRUE) %>%
  t() %>%
  prcomp()

# Quick inspection to check whether first two PCA components capture most of the variability
summary(k10pca)$importance %>% pander(split.tables = Inf)

# Plot PCA
k10pcaPlot <- k10pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k10dge$samples) %>%
  ggplot(aes(PC1, PC2, colour = group, label = rRNA)) +
  geom_point(alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  labs(
    x = paste0("PC1 (", percent(summary(k10pca)$importance[2, "PC1"]), ")"),
    y = paste0("PC2 (", percent(summary(k10pca)$importance[2, "PC2"]), ")"),
    colour = "Genotype",
    title = "k = 10"
  ) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

k10pcaRrna <- k10pca$x %>%
  as.data.frame() %>%
  rownames_to_column("sample") %>%
  as_tibble() %>%
  dplyr::select(sample, PC1, PC2) %>%
  left_join(k10dge$samples) %>%
  ggplot(aes(PC1, rRNA, label = rRNA)) +
  geom_point(aes(colour = group), alpha = 0.8, size = 3) +
  geom_text_repel(show.legend = FALSE) +
  geom_smooth(method = "lm") +
  labs(
    x = paste0("PC1 (", percent(summary(k10pca)$importance[2, "PC1"]), ")"),
    y = "rRNA proportion",
    colour = "Genotype",
    title = "k = 10"
  ) +
  scale_y_continuous(labels = percent) +
  scale_colour_discrete(labels = c("Wildtype", "Mutant"))

Differential expression

k10design <- model.matrix(~rRNA, data = k10dge$samples)
k10voom <- voom(k10dge, design = k10design)
k10fit <- lmFit(k10voom, design = k10design)
k10eBayes <- eBayes(k10fit)

k10topTable <- k10eBayes %>%
  topTable(coef = colnames(k10design)[2], sort.by = "p", n = Inf) %>%
  set_colnames(str_remove(colnames(.), "ID\\.")) %>%
  rownames_to_column("mer") %>%
  mutate(BY = p.adjust(P.Value, "BY")) %>%
  mutate(DE = BY < 0.05) %>%
  dplyr::select(
    mer,
    AveExpr,
    logFC,
    P.Value,
    FDR = adj.P.Val,
    BY,
    t,
    DE,
    everything(),
    -B
  ) %>%
  as_tibble()

Distributions

ggarrange(
  k5dist, k6dist, k7dist, k8dist, k9dist, k10dist,
  ncol = 2, nrow = 3, common.legend = TRUE, legend = "bottom" 
)

Distributions of kmer counts for each value of \(k\)

PCA

ggarrange(
  k5pcaPlot, k6pcaPlot, k7pcaPlot, k8pcaPlot, k9pcaPlot, k10pcaPlot,
  ncol = 2, nrow = 3, common.legend = TRUE, legend = "bottom"
)

Principal component analysis for all values of \(k\). As \(k\) increases, WT and mutant groups start to separate.

ggarrange(
  k5pcaRrna, k6pcaRrna, k7pcaRrna, k8pcaRrna, k9pcaRrna, k10pcaRrna,
  ncol = 2, nrow = 3, common.legend = TRUE, legend = "bottom"
)

Principal component analysis for all values of \(k\). As \(k\) increases, WT and mutant groups start to separate.

Clustering

ggarrange(
  k5den, k6den, k7den, k8den, k9den, k10den,
  ncol = 2, nrow = 3, common.legend = TRUE, legend = "bottom"
)

Hierarchical clustering of samples based on kmer counts. As \(k\) increases, WT and mutant groups start to cluster.

Differential expression

topRes <- function(x, cap){
  x %>% 
    dplyr::select(mer, AveExpr, logFC, P.Value, FDR, BY, DE) %>%
    mutate(
      AveExpr = format(round(AveExpr, 2), nsmall = 2),
      logFC = format(round(logFC, 2), nsmall = 2),
      P.Value = sprintf("%.2e", P.Value),
      FDR = sprintf("%.2e", FDR),
      BY = sprintf("%.2e", BY)
    ) %>%
    dplyr::slice(1:100) %>%
    datatable(
      options = list(pageLength = 10),
      class = "striped hover condensed responsive",
      filter = "top",
      caption = cap
    )
}

k = 5

topRes(k5topTable,
       cap = paste(
         "The top 100 differentially expressed 5-mers.",
         nrow(dplyr::filter(k5topTable, DE)),
         "of",
         nrow(k5topTable),
         "detected sequences were classified as DE with BY p-value < 0.05."
       )
)

k5topTable %>%
  ggplot(aes(P.Value)) +
  geom_histogram(
    binwidth = 0.02,
    colour = "black", fill = "grey90"
  ) +
  labs(title = "k = 5")

Histogram of p-values. Values follow the expected distribution when there are many differences.

k5topTable %>%
  ggplot(aes(logFC, -log10(P.Value), colour = DE)) +
  geom_point(alpha = 0.5) +
  scale_colour_manual(values = c("black", "grey50", "red")) +
  geom_text_repel(
    data = . %>%
      # dplyr::filter(-log10(P.Value) > 4 | logFC > 4 | logFC < -2),
      dplyr::filter(-log10(P.Value) > 4 | logFC > 3 | logFC < -2.5),
    aes(label = mer, color = "black")
  ) +
  labs(x = "logFC", y = expression(paste(-log[10], "(p)")), title = "k = 5") +
  theme_bw() +
  theme(legend.position = "none")